The Telemac system
From small scale processes…
…. to large scale applications
SISYPHE
DELWAQ (Deltares)
Free surface flow
Sediment transport
Water quality
Waves
TELEMAC-2D
TOMAWAC & ARTEMIS
SEDI-3D
TELEMAC-3D
2 dimensions
St Venant 3 dimensions
Navier-Stokes
Libraries, pre- and post-processors
BIEF Blue Kenue
FUDAA
ParaView Finite elements
User club (October 2011): 130 participants
2000 users in more than 100 Countries
www.opentelemac.org
Parallelism with domain decomposition
All options parallel (including the method of characteristics!)
With one million points : speed-up of 16 between 64 and 1024
processors, on IBM Blue Gene
Tested up to 8096 processors
Multi-scale, multi-physics
Coupling of models
Chaining
Internal coupling
Telemac-2D or Telemac-3D/Sisyphe
Telemac-2D/Tomawac
Platform (Open MI)
Laboratoire National d'hydraulique et Environement
Hydraulic Sediment transport
Sediment in estuaries
Salinity intrusion 26 décembre 2004 0h58 TU
Tsunamis
propagation
Wave and tide in the Atlantic
Ocean
Bief Library
Laboratoire National d'hydraulique et Environement
Unstructured grids
FORTRAN 90, PERL, MPI
Finite elements / Finite volumes
Implicit schemes
Fundamental operation on matrix, vectors, scalars
0
0
0 0
0
0
1
- Definition of basis function:
- Decomposition of each variable:
- Variational principles:
- Mass matrix:
n
jjjff
1
00 MFf
dM jiij
11
n
jj
Telemac-2D / Telemac-3D
Riad Ata Chi-Tuàn Pham
222*
/30logvu
ku
s
RANS
UUUU
U
)(
0
tpgt
),,( wvuU
x
U h
z
2
2
1UCDb
sk
hLog
u
U 111
*
dzvuh
VU ,1
,
hh
hh
Zgt
sbs
ττDUU
U
1
0
Uh
t
h
Turbulent flow
Hydrostatic
non-hydrostatic pressure
Bottom friction
Shallow water (St venant)
Logarithmic velocity profile (first plan)
Quadratic frition law
Logarithmic velocity profile
h<<L
Hydrostatic pressure
Non recirculating, well mixed
Equations
Other Important features
Cartesian or spherical coordinates
Courant numbers up to 10
Robust and efficient
CSDMS, Oct 2011
Telemac-2D / Telemac-3D
Sensitivity of model results
Turbulence model (k-e,mixing length)
Vertical grid resolution
Surface libre
Noeuds du maillageFond
Surface libre en tant que somme du fond et de la hauteur
Treatment of dry zones
No element removal, all points treated even if dry
Continuity, positivity of depth, conservation and monotonicity of tracers ensured by
an edge-based treatment of fluxes (Hervouet et al., AIRH congres, 2011)
Malpasset dam break
Malpasset dam, 48 million m3,
broke on 2 December 1959, there
were 433 casualties.
Mesh
26 000 elements
DT=4s
100 DT
Malpasset dam break
0
10
20
30
40
50
60
70
80
90
100
0 2000 4000 6000 8000 10000
model
Telemac-2d
Distance from the dam (m)
Maximum elevation (m)
Linux
HP Z600
Telemac-2D Telemac-3D
(2 planes)
1 proc. 52 s 188 s
8 proc. 11 s 36 s
Third generation spectral wave model F: variance density directional spectrum
Conservation of the wave action: Shoaling
Refraction
Non-linear interactions
Wind generation
Wave dissipation (breaking, white capping)
Wave-current interaction (Telemac-2d/Tomawac)
Applications: oceanic to coastal zones
Tomawac
Giovanni Mattarolo
Waves in the Atlantic Ocean
Processors CPU time (s)
1 90 000
8 18 000
16 6 500
20 5 500
24 6 000
Oceanic wave model
25 500 elements
DT=600s
1 year computation
Hs
Tm
m
Hs
Tm
m
Dec 1999 + + Observé
---- Simulé
Sisyphe / Sedi 3d
Pablo Tassi Catherine Villaret
Erosion rate
Deposition rate
Zref
• 2D suspended load
diffusionTurbulent
t
Settling
s Cgraddivz
CWCgradu
t
C))((
)()(.
h
DECK
hCU
t
Ct
1)(.
0)()1(
DE
t
Zn b
3D suspended load
0)1(
s
b Qdivt
Zn
Bed load
Total load /bed load formula
Exner equation
Cohesive
Erosion-deposition laws
(Krone and Partheniades)
Consolidation model
Non cohesive
Erosion-deposition rates
Equilibrium concentration formula
(e.g. Van Rijn, 1984)
• Deposition rate (implicit):
• Sand grading effects
* 5 1.2 2(1.83 10 . ) / ²
0.0002 / ²
e s jC N m
M kg m s
1
20
e
ME
20
1d
WsCD
Active
layer
zrefsCWD
eqsCWE
D50>60mm D50<60mm
2D suspended load
Correction on the convection
(Huybrechts, Villaret, and Tassi, River Flow 2010)
Assuming:
- Log velocity profile
- Rouse concentration profile
Sisyphe / Sedi-3D
3D suspended load
Treatment of boundary condition
Sensitivity of model results to
-turbulence model (k-e, …)
-Vertical grid resolution 1
)()(
)()(
ha
h
a
ha
dzzCdzzU
dzzCzUh
bZt
z
CE
0 inf
UCUCCU
U1 z1
Zf’=Zf +z0
Zs
z’=z -z0
z0
h
Plan fictif
V6p1
10-3
10-2
10-1
10-3
10-2
10-1
C
z(m
)
Concentration profile (=10-4m2/s)
Telemac 3Dmodified Rouse profileRouse profile
*
*
2/
u
W
h
c
uz
zhCC
: laminar diffusivity Finite values
infinite
finite 0 0.2 0.4 0.6 0.8 1 1.2
10-3
10-2
10-1
100
velocity U(m/s)
z(m
)
Telemac 3Dlog profile
van Rijn (1987)
h= 0.39 m
U= 0.51 m
d50=0.16 mm
Trench evolution in 2D/3D
3D model
16 planes
non-hydrostatic
pressure
concentration profiles (tm1=k-e; tm2=mixing length)
Laboratoire National d'hydraulique et Environement
2D model overestimates transport rates
Convection velocity correction
Improves the results
2D model : good compromise between model accuracy and
computational time for non-recirculating flow
Sisyphe
• Data
Tel3D
Bed evolution after t=15 hs
Trench evolution in 2D/3D
Secondary flow pattern in river
Computed velocities along XS-3
• 6,623 elements x 15 layers
• time step = 0.1 s
• steady state 100,000 time steps
Telemac-3d : flow patterns at a river difluence (Tassi, Vionnet and Morell, 2011)
Meandering channels
Telemac 2D/Sisyphe
Secundary current
parameterization in 2D
Yen’s experiments (1995)
Telemac 3D/Sisyphe
Turbulence model
Onishi’s experiments (1972, 1976)
•suspension = 54% Qt
•bedload = 46% Qt
•bedload only
Analena Goll, PhD student
Numerical simulation of bedforms
Flume experiments
•Length: 30m
•Width: 5m (2x2m)
•Discharges: 77-240 l/s
•Bed-load: sand (D50≈1mm)
Numerical modelling with
Telemac3D / Sisyphe
Analena Goll, PhD student
Numerical simulation of bedforms
Compute flow using total roughness
Compute the transport using local
small-scale ripple roughness only
Flow based
on total bed
roughness
Total bed
roughness
Telemac2D
Sisyphe (in ride.f)
Method of feedback for the bed roughness
k’s ~3d
ks ?
0 0.2 0.4 0.6 0.8 1 1.20
0.2
0.4
Current velocity (m s-1)k
s (
m)
0 100 200 300 400 500 6000
0.2
0.4
W+C Mobility Number
ks (
m)
ks ripples
ks megaripples
ks dunes
ks sisyphe
ks total
2
,2
,2
, dsk
msk
rsk
sk
bed roughness predictor (dunes +
megaripples + small-scale ripples), variable
in time and space (Van Rijn, 2007)
Laboratoire National d'hydraulique et Environement
Large scale morphodynamics
Field validation of the modelled
bed roughness (for dunes)
Dyfi Estuary (Wales,UK)
Shallow sandy macro-tidal estuary
Multiple banks and channels
Tidal range : 2 to 4.3 m offshore
Sediment : sand 0.2-0.25mm
River input : 25 m3/s annual mean
400 m3/s 1 in 100 yr
Triangular Mesh
BlueKenue 91,000 nodes,
resolution :1km / 15m
Dify estuary morphodynamics
Bathymetry
LiDAR survey (2004) +
Multibeam Multibeam(2007)
Bedforms measurements
5.0sk
A
B
Predicted ks (m)
(max values during spring tide)
VanRijn method (2007)
dune roughness in channels
ripple roughness over tidal flats
2ks, max
2ks, max,
neaps
2ks, average
raw av_15m
Dify estuary morphodynamics
E
Comparison with measurements along transect AB (m) 5.0sk
Comparison at Point E for two M2 tidal cycles
With ks feedback No feedback (ks = 0.10 m)
Effect of feedback method for bed roughness
Bottom shear stress increases (by 122%)
Total sediment transport rate increases (by 74% on average)
Flood Ebb
Dify estuary morphodynamics
Diversity of applications in complex environment:
Meandering channels, Macro-tidal estuaries, littoral applications, …
Dune formation to mesoscale morphodynamics (10-100 km)
Bed roughness predictor
reduces some of the uncertainty in model results
avoids possible unconsistency between hydrodynamics and sediment
transport
More physical processes
Sand grading algorithm and mixed sediments
Mud consolidation and flocculation
Conclusion
Uncertainty in the sediment transport models >>> hydrodynamics
models
New validation test cases
Uncertainty analysis (automatic differentiation)
Jean-Michel Hervouet
Emile Razafindrakoto